Tackling illegal mining with tech

0:00Welcome to Tech Life on the BBC World Service, the programme about technology and how it's changing all our lives. I'm Chris Valance. This week we're looking at the problems caused by illegal gold mining in Ghana. We'll hear how drones, AI and GPS trackers are being used to tackle the trade that's causing damage to the environment and polluting water supplies. That's in a moment. Also, Meta goes to court as it battles ads for so-called nudifying apps. And the pay-as-you-go cooker that's winning awards and improving health in Africa.

0:55We start today in Ghana, in West Africa. Agriculture and mining are central to the country's economy. Gold mining is big business. with the price of gold soaring in the last two decades. Ghana is Africa's largest producer of the precious metal and the sixth largest producer in the world. Most mining is legal, but illegal gold mining, which is called gallumsane, Ghana, is a serious problem. And recent estimates suggest around 70 % to 80 % of small mines are unlicensed. Illegal mining tears up the land and damages water supplies, harming crops. But it also damages the health of those who do the digging.

1:38BBC News reporter Thomas Nardi visited an illegal mining area in Ghana late last year. In this clip, a miner tells him about using mercury, which is toxic, to separate gold from soil. The miners here tell me they mostly mine on cocoa and rubber plantation lands. They either purchase a acre of land for around$2 ,000 or they work for the landowners. Galamse is difficult. It's basically digging of the soil and trying to separate the gold from the soil. We use mercury to separate the gold from the soil. It's the mercury that is highly toxic for the environment and if inhaled or consumed by a human, it could be fatal.

2:22That's Thomas Nardi reporting there. Now technology is being used to combat the illegal mining trade with drones, AI and GPS trackers at the heart of the government's strategy. Dr Sylvester Akpa is a lecturer at the University of Mines and Technology in Ghana. He's also a consultant in the use of this technology to Ghana's Minerals Commission. He's been telling me about their drones programme. Presently, the Minerals Commission has over 25 long-range drones. We have trained about 50 drone pilots within the commission. So what we do specifically is that Minos Commission has what we call a database, right?

3:02We call it the Mine, Cadastre and Administrative System that has coordinates of all licensed small scale mining sites within the country. Okay. Once the drone is flying and then we are capturing objects of interest. And in this point, the objects of interest are what? Excavators, mudded pumps, what we call sluice boards, human beings, shelters and all this. So once the drone is capturing this, all this live feed is brought into the control room where we have set up in Accra. And then the AI models comes to play. It detects these objects of interest for us and automatically compares the coordinates where these objects of interest are in live video.

3:44And then automatically pairs it with the coordinates in the mind cadastro administrative system. So if the coordinates of the objects matches with what we have in the mine cadastro system, automatically we are saying that these objects are on a licensed mining flag. So essentially the drone is sending you back images. The AI is going through that, spotting things like muddy ponds, mining equipment, all the other stuff, the buildings, the people associated with small-scale mining, sends it back. The AI spots it. But the AI then compares the coordinates of what is found with the coordinates of registered mines in your mine database.

4:27And if there isn't a match, that's an illegal mine. That's right. That's right. So what happens when you've got a flag? What's the next step? This is what happens. So once we actually get a flag, then it means that Minerals Commission has district offices across virtually all the mining areas. okay so the the coordinates that is flagged is automatically sent to the closest mining district offices with the geolocation of the area where that flag that equipment has been flagged so and within a very short period of time that these offices are able to move into those areas and be able to probably make arrests or id out the equipments that has been flagged and then the Equipments are confiscated for further sanctions or other things like that.

5:14And this also leads to the mines being closed and all of that. As well as the drones, you're also tracking the tractors and the diggers and the sort of the heavy machinery that the illegal miners are using. And you're doing that using GPS. Tell me about that. Yes, so we are tracking basically all heavy mining equipments, especially excavators and bulldozers. and this is what we are doing we actually install pre-programmed or customized trackers in them and more so what we are doing is that we are approaching it from two angles there are brand new excavators that are getting into the country every day right so we are tracking them from the port of entry and by this we're actually getting the details of the owners who imported them we're getting the details of the equipments that are there so that at any point in time we know where they are we know where where they are moving to and all that there are already excavators or earth moving equipment within the country okay it's estimated that presently we have over 15 000 earth moving and mining equipment in ghana alone that are involved in mining so it is actually a very huge tax but then we are on it we're actually tracking all of them to ensure that we know where exactly they are who owns them and then be able to monitor their day-to-day activities within the mining district on a daily basis and in real time.

6:39And you're able to stop them, if you like, working outside of the area they're meant to be in? Exactly. We are tracking a lot of equipment. So the trackers have been pre-programmed, okay? And based on the coordinate system that we are using, so we paired equipment to the site with the coordinates. So at any point in time, once you leave your coordinate, your designated coordinate system, automatically the machine goes off on its own. So we don't have human beings sitting down to shut them down. No, we don't do that. So the software actually takes charge and shuts the equipment down automatically once you leave your defined concession area.

7:17So this is geofencing, as it's called. Effectively, that's like if you like an invisible boundary beyond which the bulldozer or the excavator can't go, it'll just stop working. Exactly. So we geofence each mineral right area or with each concession area at any point in time and then we pair the equipment to that concession area. So at any point in time, we have eyes on the movements of these heavy mining equipments. That's Dr Sylvester Akpa, a consultant to Ghana's Minerals Commission, speaking to me from Accra. And Dr Akpa also told me he's sharing his knowledge with neighbouring countries in Africa, which also have illegal mining problems.

8:06So-called nudifying apps, as the name suggests, create explicit images that simulate removing clothes from pictures of real women, deeply distressing for those targeted and widely regarded as a form of AI-enabled abuse. Now Meta, which owns Facebook and Instagram, says it is taking legal action against Joy Timeline HK Ltd, an entity it says is behind the nudifying app Crush AI, to stop it advertising on its platforms. Meta says it discovered thousands of ads for the app that broke its terms of service. Now it's sharing information with other tech firms to help them take action, and is also deploying new technology to identify and remove nudifying app adverts.

8:55These new steps follow calls from researchers for more action to be taken. Alexius Manzales, director of the Security, Trust and Safety Initiative at Cornell Tech in the US, wrote in January about the proliferation of ads for crush AI on meta platforms. So what did he think of their latest efforts to combat nudifying apps? I think the move is great. It's welcome, but it's late. It has been more than a year that myself and others have been flagging Crush AI as a sophisticated actor that got thousands and thousands of ads past Meta and served to millions of its users around the globe. Still, I'll take it.

9:38It's good news. Tell me a little bit about Crush AI. Yeah, so it's a network of websites that started off as allowing the users to turn any photo into a nude version of that same photo. And as we know, this is an epidemic globally. Both the UK and the US have seen hundreds of thousands of these cases self-reported, including in schools, so targeting teenagers. And Crush AI has been very effective at getting around Meta's moderation by creating dozens of different websites, all pointing to the same service, hundreds of different account advertisers' account, and thousands of different assets. So it's certainly not the only nudifier.

10:25There's many and some are even larger, but it has been a persistent one and a sophisticated one in getting around Meta's rules. What do you want to see these firms do to get rid of these kind of adverts more quickly? The platforms have been slow and not chosen to kind of throw the all that they have against them. I think Meta's decision to take legal path is a sign that they are stepping up and ramping up. The other news was that they've developed a custom classifier to flag this content because even though the nudifiers turn photos into nudes, those nudes tend to be shared privately in WhatsApp groups and Snap messages.

11:10So it is harder to find these than with an old-school nudity classifier. But again, what I want to see is a serious commitment at the highest levels from these companies. Unfortunately, even as Meta was announcing this, I found more than 100 ads for Crush.ai and other nudifiers. And Meta is not alone. We see these services have apps on the Apple and Google Play stores. We see that they have channels on Telegram with more than 100 ,000 followers. We have verified accounts on X. So there's all kinds of mainstream infrastructure that need to step up their game and just staff up the teams fighting on this.

11:56Look, I was a trust and safety in a platform. I know this stuff can be difficult and requires resources. The CEOs should be giving the teams resources. Can they not use AI to spot these ads? I mean, you know, you, I and pretty much everyone else with a pair of eyes, I'm sure, could say, well, look, that's a nudifying app. Doesn't sound like the hardest challenge in the world to present to an AI to say flag things that look like nudifying apps. look we've been told for more than a year that ai will replace humans in a variety of roles i'm still waiting i'm still waiting and over the past year i have kind of almost manually and through burner accounts found the 10 000 plus ads for crush ai these things as we see require attention to circumvention require an understanding of what's actually the end goal certainly AI can help but AI has clearly failed at replacing or enabling kind of the human eye in detecting and taking down the harms that by the way ironically AI is causing right we wouldn't be in this mess we wouldn't be in a situation where six percent of American teens claim that have been targeted by a deep fake nude if it wasn't for widespread access to cheap AI enabled notification.

13:20So this is a mess that AI has caused, and it's not a mess that AI has fixed. Alexios Monsalos there. Well, we approached Joy Timeline HK Ltd. for comment, but they have not yet responded.

13:43You're listening to Tech Life on the BBC World Service, with me, Chris Valance, and, of course, with you, our listeners. And talking of listeners, hello to Tom Schroeder, who messaged us from San Diego in the United States. He got in touch because we mentioned his city's weather in last week's show, which Tom tells us has so far been June gloom. Well, just for you, Tom, as an experiment in hyper-personalised weather forecasting, we've looked up San Diego on the BBC Weather website and you should have warm sunshine for the rest of the month. Thanks for getting in touch and stay classy. Also in the US, Patrick Wiggins from Utah emailed us after hearing last week's report about electric cars.

14:28Since 2015, he's clocked up over 263 ,000 kilometres of mostly trouble-free electric driving. Patrick says the only required maintenance has been a door handle, one air filter and a couple of 12-volt batteries. The big batteries that power the wheels remain fine with very little capacity loss. And talking of tech that keeps on trucking, Steve Gustavuson emailed from San Francisco in the USA. Steve says, Hello, I listen to your show and BBC World Service Radio on my iPod Touch, seventh generation, circa 2022. Keep up the great reporting. Nice to hear from you, Steve. I've still got an iPod Classic, now about 20 years old, and though it mostly lives in a drawer, it does have one great benefit over more up-to-date alternatives.

15:17No one can call you or text you on it. And if you want to tell us about the one bit of tech you really can't do without, please do get in touch. Our address is techlife at bbc.co.uk or send us a WhatsApp text or voice note. The number is plus 44 330 1230 320. I'll repeat the details at the end of the programme. And still to come, maybe you have a pay-as-you-go phone, but one company hopes their pay-as-you-go cooker will transform lives and health in Africa.

15:58Now, what's the hardest thing to say to someone? I'm sorry, it's not working out, you're fired. Well, for some generative AI systems, it's apparently, I don't know. So-called large language models, the kind of AI that powers popular chatbots, are prone to making things up, hallucinating as it's known, when they don't have the data to accurately answer a question or a prompt. And as we rely on AIs more than ever, these hallucinations can have serious consequences. But unlikely AI, a London-based business, thinks it has the answer. The company's founder, William Turnstall-Peddo, was one of the minds behind Amazon's Alexa.

16:44TechLife's Graham Fraser spoke to William recently and began by asking him to outline the problem of hallucination. The magic technology behind ChatGPT is called an LLM, large language model. And what it does is, in an incredibly sophisticated way, it predicts words. So it predicts language. And most of what it predicts is true. but the way it generates language also means that it invents completely plausible untrue things as well as an almost an intrinsic part of the way it operates. So a percentage of what this technology produces is incredibly plausible fiction, things that people believe but turn out to be completely untrue.

17:23And of course for many different reasons that can be quite dangerous when it's making up answers. if these answers are used in journalism or if they're answers used in university research or for a whole plethora of different reasons. It's quite a challenge if the AI models, if we're trusting them and then what they're saying isn't true. Yeah, and it's actually almost the worst user experience. If you ask a system a question and it says, I don't know, that's obviously much, much worse than giving you the right answer, but at least you don't learn something that isn't correct. but often this technology produces something that is exactly what you want to hear that's incredibly plausible that sounds like it's true but actually isn't.

18:05So is this what you are arguing that actually the answer to hallucination is AI models that say I don't know when they don't know? That would definitely be better than inventing something. So yes exactly right. So what we're working on as Unlikely AI is addressing that problem and we're addressing that with deep tech that combines the magic of LLMs with other types of technology to create an experience where if it produces an answer, it's accurate. If it produces an answer, it's fully explainable. And if it isn't sure, it will say, I don't know, rather than inventing something. So obviously in the chatbots that many people at home who are listening will use are dominated by the big tech companies, the likes of Google.

18:50Obviously you've got ChatGBT from OpenAI. What would you think if those huge models, if they could adopt the same principle of I don't know, how do you think that would change the whole conversation around AI? These big AI labs are definitely working on this problem. It's a major problem with the industry. It's definitely something they're extremely aware of. It's something they measure. Their tech approach doesn't typically solve it, and we are solving it a different way. The broad approach we're taking is called neurosymbolic. It's a combination of neuro, which is the statistical machine learning model, and symbolic, which is a different type of software.

19:29So if you think about your spreadsheet, your spreadsheet doesn't hallucinate. Your spreadsheet will always add up two cells correctly. So we are combining that reliable class of software, that symbolic traditional class of software, with the magic of LLMs to create that experience where you can get the capabilities but also have the trust. That's William Tunstall-Peddo from Unlikely AI, speaking to Tech Life's Graham Fraser. And do let us know if you've experienced a hallucinating AI. Did it undermine your faith in the tech or was it just a minor irritation? Let us know. Our contact details will be coming up again shortly.

20:08Now, how do you cook dinner? Last year, the World Health Organization said just over 2 billion people still cook using open fires or inefficient stoves fuelled by charcoal, wood or crop waste. Burning the fuels causes air pollution inside homes, and each year 3.2 million people die prematurely from illnesses linked to cooking-related smoke. Mero Anjo-Roe lives in Kenya and has a family of five. She plants vegetables for a living and she told me how her health had been affected by cooking with a traditional stove. I used to use the traditional charcoal firewood and I used to be sickly. Why were you sickly?

20:56Was it the smoke from the liquor? Yeah, the smoke. Because of the smoke and the carbon from the charcoal. I used to sneeze, cough, having headaches. I was the one who was cooking. I used to be sickly always. Not a great thing to happen when you're looking after a family of five. Well, a company based in Kenya has won the Ashton Award for Outstanding Achievement for their work producing and supplying affordable and emission-lowering cooking stoves. Their latest innovation is an electric cooker for areas of Africa already linked to the electricity grid. Connected to the internet, the Made in Africa device has pay-as-you-cook technology.

21:41Meredith Muthoni is head of electric finance at Byrne Manufacturing. She told me they were trying to tackle three problems. So you find there's lots of smoke from traditional forms of cooking that are really harmful to the health of people and their families. It is very expensive because you'd find they'd have to buy charcoal. And as we go along, charcoal is getting more and more expensive, especially in urban areas. And another thing is the time it takes. So it's not a fuel that's available at home. So you have to walk either to the nearest market. And for rural families that use wood, they have to walk for hours to collect the firewood they need for cooking.

22:22So the effects of time, health and also how much expensive and savings it costs them to cook are really the downsides of using traditional forms of cooking. And how does burn manufacturing try and address that problem? Yeah, so at burn manufacturing, we address that problem by ensuring that we develop fuel efficient stoves across different categories of the fuel ladder. I'll put it that way. So we have the most fuel efficient wood stove, charcoal stove. And we recently launched our electric cooking products that have really transformed the lives of very many people like you've just heard. So tell me about your latest electric induction stove.

23:08Yes. So it's a single burner electric stove that's targeted to low income urban and peri-urban households that are connected to electricity. It's a bit of a conundrum. Many people think that Africa doesn't have enough electricity just for lighting, but it's there. So we have 600 million people already connected to the grid. That's about 150 million households that can access electric cooking. Though you'll find in these urban and peri-urban areas in most countries, they're still relying on charcoal or firewood as their base for cooking because these solutions have not been available to them. So what we did at Burn is engineer and design a very, I would say, market focused and based induction stove that meets the needs of these communities.

23:59And at the same time, we put it on a pay as you cook model to make it more affordable. So essentially what you're doing is you're getting rid of the sort of the big bill at the start in terms of buying the cooker. But you're spreading that bill out over payments that also cover the sort of, if you like, the pay as you go electricity as well. exactly does that still work out cheaper though for for people to do that it does and if the thing you find in most african markets the electricity cost is not significantly high i think ken is one of the ones with one of the highest ones but it's very affordable to cook on electricity compared to other forms of fuel like even buying charcoals that's what we've seen how do you do the pay as you go pay as you cook if you if you like um billing is that done through an app or I mean how is that managed?

24:48So we do have an app but we also have ussd options so there is also that we've integrated with lots of the local telco companies to make sure that it's as accessible as possible so as long as you have a mobile phone even if it's a feature phone you're able to pay for the device. And where are the stoves manufactured? So we do manufacture them we do assemble them rather in in Kenya and do have other assembly lines in Tanzania and Ghana as well. Right now we're doing mainly local assembly. We're not yet moved to full manufacturing, but that's the planning goal for the next three years. And thanks to Meredith Mathoni from Byrne Manufacturing in Kenya for talking to us.

25:34Well, we're almost done for today, but we're already cooking up ideas for next week's edition of tech life. And as ever, we'd like your thoughts and comments to throw into the pot. Is I don't know the solution to AI chatbot hallucinations that we've been waiting for? And we'd love to hear more examples of tech you simply can't do without, whether in your work or your home. Our email address is techlife at bbc.co.uk, or you can WhatsApp us on plus 44 434 330 1230 320. Tell us your name and where you live too. Today's Tech Life was produced by Tom Quinn, edited by Monica Soriano and presented by me, Chris Vallance.